- Uncovering AI
- Posts
- 🦋 Anthropic just shipped what it spent months fearing
🦋 Anthropic just shipped what it spent months fearing
Meta says America loses the AI race without these workers (and they're not coders)
In partnership with
My fellow AI explorers
Two months ago, Anthropic looked at its most powerful model and decided the world wasn't ready for it. This week, they handed a version of it to everyone with a Claude login. Same brain, different leash. That tension, between "this could change everything" and "this could be dangerous in the wrong hands," is the whole story of AI in 2026, and it's sitting right at the top of today's edition.
In today’s edition:
🦋 Anthropic releases Claude Fable 5, its most capable public model ever, with a built-in safety net
🔌 Meta's president says America can't beat China in AI without welders and electricians
⚖️ AI hallucinations get lawyers banned, China runs "quiet" layoffs, and Apple finally ships
Six people doing the work. Your headcount is one.
Your finance close runs in #finance. Stripe and QuickBooks reconciled, runway updated, posted Sunday night without you asking.
Engineering review lands in #eng. Viktor pulled the open PRs, left comments on auth-refactor, flagged a dependency blocking api-pagination.
Campaign brief lands in #growth: Meta CPA up 18%, recommendation to pause broad match, a draft landing page already deployed for the variant test.
You hired him on day zero. He lives in Slack and Microsoft Teams alongside your contractors and investors, connects to 3,000+ tools, pushes back when you ship something dumb.
"Viktor is now an integral team member, and after weeks of use we still feel we haven't uncovered the full potential." Patrick, Director, Yarra Web.
AI Competition
🔌 The AI Race Will Be Won by Electricians, Not Engineers
Here's a sentence I didn't expect to write: the biggest threat to America's AI ambitions might be a shortage of welders.
Meta President Dina Powell McCormick went on Fox Business this week and said it plainly: America can't compete with China in AI without the skilled tradespeople who physically build the data centers. To do something about it, Meta is teaming up with Mike Rowe's foundation to launch America's Workforce Academy, a $115 million training program.
What's on the table:
🛠️ Paid training with a job at the end. Participants get paid while they train, earn industry-recognized credentials, and are guaranteed a job opportunity on completion, all without quitting their current gig.
📍 Four states first. Louisiana, Ohio, Indiana, and Texas, with plans to expand alongside the data center buildout.
⚡ The shortage is electricians, fiber techs, and welders. The trades powering the AI boom that almost nobody is talking about.
The framing is the interesting part. Tech companies are pouring billions into AI infrastructure, and they've hit a wall that has nothing to do with software or silicon. As Rowe put it, these jobs "exist out of sight and out of mind." Powell McCormick went further, calling skilled workers "the American heroes" without whom the country simply can't keep pace.
There's a quiet irony here worth sitting with.
The same AI boom that's automating away white-collar work is creating massive, urgent demand for the kind of hands-on labor that can't be automated, at least not yet. The data center has to be wired by a human. The cooling system has to be welded by a human.
AI's physical footprint is turning into one of the largest blue-collar hiring drives in a generation.
🔮 Prediction: Over the next two years, "AI infrastructure" becomes a bipartisan jobs story, not a tech story. Expect more hyperscalers to launch their own trade academies, expect federal and state incentives to follow the workforce rather than the chips, and expect the political narrative around AI to flip from "it's taking our jobs" to "it's building them," at least for the trades. The companies that lock up skilled-labor pipelines now will have a moat that no amount of GPU spending can buy.
If you've worked a trade, would a Meta-backed academy with a guaranteed job get you into a data center? I want to hear from the people actually doing this work, reply and tell me.
AI Tools
🦋 Anthropic Shipped the Model It Was Afraid Of
Anthropic just released Claude Fable 5, the most capable model it has ever made available to the public. And the framing tells you everything you need to know: they're equal parts proud and nervous about it.
Here's the setup. Anthropic now has a tier of models it calls "Mythos-class," which sit above its Opus models in raw capability. The first one, Mythos Preview, was so good at finding and exploiting software vulnerabilities that the company kept it locked away for cyber-defenders only.
Fable 5 is that same underlying model, made safe enough for the rest of us through a layer of guardrails. Mythos 5, the unrestricted twin, stays behind the velvet rope.
What you need to know:
🧠 State of the art on nearly every benchmark. In early testing, Stripe reported Fable 5 ran a code migration across a 50-million-line codebase in a single day, work that would have taken a team over two months by hand.
🚧 It has an automatic off-ramp. When you ask about cybersecurity, biology, chemistry, or model "distillation," your request quietly gets handed to the weaker Claude Opus 4.8 instead. Anthropic says this happens in under 5% of sessions.
💰 The pricing is aggressive. Ten dollars per million input tokens, fifty per million output, less than half what Mythos Preview cost. It's free on Pro, Max, and Team plans through June 22, then it moves to usage credits while capacity catches up.
The detail that should stop you: the name itself is the safety story. "Fable" and "Mythos" come from the same root, "that which is told." The only difference between the model you can use and the one you can't is the guardrails. Anthropic is essentially admitting that what the model is capable of make it a risk, not just its application.
And the risk is real. The earlier Mythos Preview reportedly found and exploited zero-day flaws in every major operating system and browser when directed to, including a 27-year-old bug in OpenBSD, an OS famous for being hard to crack. That's why Anthropic ran a 1,000-hour external bug bounty hunting for jailbreaks before letting Fable out the door.
🔮 Prediction: This is the moment "capability" and "access" officially split into two separate products. Expect (and hope) the entire industry to follow Anthropic's lead: one frontier model, sold at two or three different guardrail levels depending on who you are and what you can prove about yourself. The era of everyone getting the same model is ending. Your AI access is about to come with a background check.
Would you trade a more powerful model for one that occasionally refuses to help you, or ties one hand behind its back? Hit reply, I'm genuinely curious on this one.
30-Second AI Play
Turn Any Webpage Into a Personal Watchdog With Apple's New Safari
Apple just previewed a feature that quietly kills one of the internet's most annoying chores: refreshing a page over and over, waiting for something to change. It's called Notify Me, part of the next generation of Apple Intelligence, and you can use the exact same logic today with any AI assistant while you wait for the fall rollout. Here's the workflow:
Pick the thing you're waiting on. A concert restock, a price drop on a flight, a product back in stock, a job posting going live. Anything that lives on a single webpage.
Copy the URL of that exact page, not the homepage. Specificity is everything here.
Hand it to your AI assistant with a precise instruction: "Check this page once a day and tell me the moment [X] changes." Be explicit about what "change" means to you, in stock, under $200, new date added.
Set the cadence. Daily for slow-moving things, hourly for drops and restocks. Don't over-poll, you'll just create noise.
Define the payoff. Tell it exactly what to send you when the trigger hits: a one-line alert, the new price, a direct link to checkout.
💡 Pro tip: This is most powerful for the high-stakes, time-sensitive stuff where being first matters, limited drops, repricing, restocks. When Apple ships Notify Me natively this fall, you'll be ahead of everyone still hitting refresh by hand.
Other Relevant AI News!
⚖️ A federal judge in Mississippi just sanctioned all four lawyers in a single case, on both sides, after each independently filed briefs full of AI-fabricated citations, with two attorneys banned from the court for two years.
🇨🇳 While the West fights over AI layoffs, China is doing them quietly. Firms are cutting contractors in small batches after deploying agents like OpenClaw, staying just under the radar of Beijing's scrutiny.
🇮🇳 Meta is going global with its compute, partnering with Reliance to build its first AI data center in India, a 168-megawatt facility in Jamnagar, as $400 billion floods into the country's AI ecosystem.
📈 Supermicro is raising $7 billion to fund a staggering $39 billion backlog of AI server orders from 20-plus customers. It’s a raw read on just how hot data center demand still is.
🍎 Apple finally shipped real AI, a new Siri, photorealistic image generation, a Passwords tool that agentically fixes your weak logins, all built on a foundation model co-developed with Google's Gemini.
🤖 In Denver, an AI-powered robot is now sorting recycling 60-70 times a minute at a Republic Services plant, and the GM's verdict was blunt: "This AI thing is going to be the future."
Golden Nuggets
🦋 Anthropic split capability from access: Fable 5 is its strongest public model, with a built-in fallback for risky topics.
🔌 Meta says the AI race runs through skilled trades, and is spending $115M to train the workers who build the data centers.
⚖️ The AI hallucination reckoning hit the courtroom hard this week, with lawyers on both sides of a case sanctioned at once.
Would love to hear your thoughts! Send me your thoughts by replying to this email (yes, I read them all :)
Until our next AI rendezvous,
Anthony | Founder of Uncover AI